When your product is an agent that acts autonomously, bad customer feedback doesn't just waste your time — it steers the automation wrong, and the mistakes compound at machine speed.
Rob Fitzpatrick's The Mom Test is a short, merciless book about why most customer conversations produce garbage data. The culprit is the founder's own framing: ask someone what they think of your idea and social pressure does the rest, generating warm praise that predicts nothing. The fix is to stop talking about your product entirely and instead dig into how people actually spend their time, what has already cost them money or sleep, and what workarounds they built before you showed up.
The stakes shift when you are building an agent-native product because you are not just validating a feature — you are mapping a workflow you intend to hand to software that will run unsupervised. A user who says your agent sounds great has told you nothing useful. A user who walks you through the last time that process failed, names the hour it cost them, and shows you the spreadsheet they built to compensate — that person has handed you a spec. Commitment signals matter even more here: if they will not change one habit or share one real example, the agent has nowhere to land.
- Ask about the last time the problem actually hurt, not whether your solution sounds appealing
- treat every unprompted workaround as a map of where the agent should operate
- treat enthusiasm without behavioral change as a warning, not a green light.
